Robot Perception

What it does

Gives Claude a battle-tested playbook for robot perception systems.

• Sensor landscape: comparison tables of RGB, stereo, structured light, ToF, spinning/solid-state LiDAR, IMU, force-torque, tactile and event cameras (output shape, range, rate, best use), plus a hardware-to-SDK-to-ROS2-package mapping for RealSense, ZED, OAK-D, FLIR/Basler, Velodyne, Ouster and Livox.
• Three calibration workflows: checkerboard intrinsics with coverage tracking, sub-pixel refinement and per-image reprojection diagnostics; stereo and camera-to-LiDAR extrinsics via stereoCalibrate/solvePnP; eye-in-hand and eye-to-hand calibration via cv2.calibrateHandEye with a verification routine.
• Quality checklists: target RMS thresholds (<0.5 px intrinsic, <1 px stereo, <5 mm hand-eye), minimum image/pose counts, and explicit triggers for recalibration (bumped camera, focus change, thermal drift).
• Streaming architecture: dedicated capture thread, bounded deque with drop policy, timestamping at capture, and FPS/drop diagnostics.
• Synchronization: timestamp nearest-neighbour soft sync, RealSense inter-cam hardware sync, PTP for GigE cameras.
• RGB pipelines: precomputed undistortion maps, detection wrapper with workspace filtering and track stability, median-window depth back-projection to 3D, AprilTag/ArUco pose estimation.
• Anti-patterns: same-thread capture+inference, unbounded buffers, sleep-based frame timing, timestamping at processing time.

Who it's for

• Robotics engineers wiring cameras and LiDARs into ROS2 stacks
• Developers who need hand-eye calibration for manipulation
• Teams debugging frame misalignment or latency in multi-camera rigs
• ML engineers deploying vision models on robot/edge hardware

Example uses

1. "Write an eye-in-hand calibration script for a RealSense D435 on a UR5" → pose-collection loop, calibrateHandEye call, and a sub-5 mm verification step.
2. "My checkerboard calibration reports 1.8 px RMS — why so high?" → diagnosis against the checklist: poor image coverage, fronto-parallel-only boards, autofocus drift.
3. "My camera runs at 30 FPS but the pipeline only hits 9 FPS" → refactor into a capture thread with a deque(maxlen=2) latest-frame policy.

Read the instructions in this file and follow them to help me:
https://raw.githubusercontent.com/arpitg1304/robotics-agent-skills/HEAD/skills/robot-perception/SKILL.md

What I want: (describe your task here)

Install the skill I found on Claude Skill Mart.
Copy the skills/robot-perception folder from the GitHub repo arpitg1304/robotics-agent-skills into my ~/.claude/skills/robot-perception/.
When it's done, tell me in one line what this skill can do.